testsAndMisc/NEXT_SESSION_PROMPT.md

Next session: the 10 remaining untested fixes/lib libs

Paste this whole file into a fresh Claude session opened in ~/testsAndMisc. It is self-contained. Do not go looking for the previous session's context.

The job

linux_configuration/scripts/single_use/fixes/lib/ has 22 libs. Ten still have no tests. Write them, to 100% line coverage, and take each one off meta/shell-coverage-allowlist.txt.

lines	lib	entry script
215	ubuntu_perf_fixes.sh	fix_ubuntu_performance.sh
202	arch_hardware.sh	optimize_arch_desktop.sh
180	nvidia_config.sh	nvidia_troubleshoot.sh
179	thorium_repairs.sh	fix_thorium.sh
161	hosts_guard_migrate.sh	migrate_hosts_guard_to_guard_lib.sh
141	hosts_guard_rollback.sh	same
126	arch_cpu.sh	optimize_arch_desktop.sh
122	arch_sysctl.sh	optimize_arch_desktop.sh
121	arch_perf_fixes.sh	fix_arch_performance.sh
80	ubuntu_perf_more.sh	fix_ubuntu_performance.sh

Suggested order: smallest first, grouped by entry script. The three optimize_arch_desktop.sh libs (arch_cpu, arch_sysctl, arch_hardware) share a harness; so do the two hosts-guard libs and the two ubuntu-perf ones.

The last handoff mis-grouped these — do not trust its "arch perf six share one entry script" claim. They span four different entry scripts, as above.

Everything you need already exists — do NOT build a new harness

fixes/lib/tests/ is a working suite: 20 test files, all green, all shellcheck-clean with zero suppressions. It has three layers:

• run_all.sh — the runner. This is the subject the gate measures through.
• lib_test_core.sh — the shared primitives. Read this first:
  • _t_eq / _t_contains / _t_lacks / _t_pass / _t_fail
  • _t_stub NAME BODY, and the three helpers that exist to keep call sites free of quoted $: _t_stub_stdin NAME (body from a quoted heredoc), _t_stub_cat NAME FIXTURE (print $DEV/FIXTURE), _t_stub_writes NAME ARGNUM CONTENT. Use these instead of _t_stub foo '...$1...', which trips SC2016 every single time.
  • _t_unstub NAME... (runs hash -r; deleting the file alone does not work)
  • _t_hide TOOL... — rebuilds PATH from symlinks to every real binary EXCEPT the named ones. This is how you test a "not installed" branch when the host genuinely has the tool. Do NOT drop PATH to $FAKE_BIN alone: that also hides grep/head/awk, and the case then passes for the wrong reason.
  • _t_calls / _t_reset_calls — what the stubs recorded.
• Two family harnesses, bt_harness.sh and arch_perf_harness.sh. Each sources the core, sources the real scripts/lib/common.sh for the production log_*/has_cmd helpers (it is inert at source time), and redefines the helpers that live in the entry script rather than any lib.

To add a family: copy arch_perf_harness.sh's shape. It is ~100 lines and the only family-specific parts are the entry-script helpers and the default stub list. jscpd already excludes **/lib/tests/**, so shared test setup carries no duplication risk.

The loop, per lib

# 1. write tests/test_<lib>.sh, chmod +x it
# 2. run the suite bare first -- much faster than the jail
bash linux_configuration/scripts/single_use/fixes/lib/tests/run_all.sh

# 3. then measure through the runner, in the jail
timeout 900s bash meta/scripts/shell_coverage_jail.sh \
  --subject linux_configuration/scripts/single_use/fixes/lib/tests/run_all.sh \
  --measure <lib>.sh --min 1 --fail-on-case-error -- ""

# 4. close the gap, repeat until 100.00%

Always measure through run_all.sh, never a single test file.

The bar for taking a lib off the allowlist

1. Measure 100.00% through run_all.sh.
2. Measure again at --min 100 so a shortfall is a non-zero exit.
3. Remove the line from meta/shell-coverage-allowlist.txt, then run the gate with no exemption and confirm exit 0:

   bash meta/scripts/check_shell_coverage.sh <path to lib>.sh
   # expect: "every checked shell library meets the coverage bar"
   

Allowlist is at 93 entries — count with grep -vc '^#', since the file also holds 13 comment lines.

Work the jail in a detached worktree

The jail fingerprints git status --porcelain and aborts with "a write escaped the jail" if the tree changes during a run — it compares a delta, so a pre-existing dirty file is fine but any concurrent write is not. Editing anything mid-run trips it, and so does a background build.

Use a throwaway worktree and copy one-way, INTO it, never out:

SP=/tmp/claude-.../scratchpad          # your scratchpad
git worktree add --detach "$SP/cov" HEAD
cp <files> "$SP/cov/<same paths>"      # never the reverse

Copying out of the worktree is how the last session silently reverted its own uncommitted harness edit, twice, and misdiagnosed it as another agent.

The instrument: FIXED, and now correct for two more shapes

Coverage = (PS4-traced lines | kcov hits) & (kcov line set - continuation lines)

continuation_lines() in meta/scripts/lib/shell_coverage_lines.py excludes lines kcov instruments but bash attributes no statement to. It now handles:

• continuation lines of a multi-line quoted argument (pre-existing);
• a closing brace plus an operator — } |, } >>file (added in d989cd7e). This alone took bt_audio.sh 96.97% -> 100% and arch_perf_report.sh 98.33% -> 100%.

Acceptance test if you ever touch it: dot_resolver_install.sh = 39/39 and rpi_nc_ca.sh = 73/73, through features/lib/tests/run_all.sh WITH the jail_args file beside it (that suite refuses to run bare). And no lib's percentage may go DOWN.

Note the arithmetic that bounds any denominator change: _measure computes covered_set = (executed|traced) & denominator, so a lib already at 100.00% cannot regress from excluding more lines. Over-exclusion is the only real risk — it inflates coverage and could promote an untested lib — so audit any new predicate by printing every newly-excluded line with its text.

The one known-unfixed artifact

bt_adapter.sh sits at 98.55% (68/69), still exempt, capped by line 97: the opening line of a backslash-continued assignment (result=$(dbus-send ... \). bash attributes that statement to the LAST line of the run, so the opening line can never be covered.

The predicate is narrower than "has a backslash". A backslash-continued plain command (apply_fix "..." \) is attributed to its FIRST line and covers normally — bt_pairing.sh has three and reached 100%. Only the assignment flavour caps a lib. Find them with:

grep -nE '^[[:space:]]*(local[[:space:]]+)?[A-Za-z_][A-Za-z0-9_]*=.*\\$' <lib>

Across all 22 libs that matches only bt_adapter.sh (1) and bt_audio.sh (2), and bt_audio still reached 100%, so this blocks exactly one line in one lib. Fixing it means teaching continuation_lines() the command-vs-assignment distinction. Not worth it for one entry unless you hit it again.

Two production bugs were found by writing tests. Expect more.

Both were dead branches that no test could have covered without noticing:

• bt_audio.sh:21 — timeout 3 _run_as_user wpctl status. timeout is a binary and cannot invoke a shell function, so it always returned 127 and the PipeWire health check restarted a healthy audio stack on every run.
• arch_perf_report.sh:36,65 — the journal-size regex required a space before the unit (4.2 G), but journalctl prints 305.5M. The size check and the vacuum were both unreachable.

When a branch looks untestable, check whether it is actually reachable in production before writing a test around it. The user's standing answer when asked has been "fix it in this commit".

Traps that cost real time

• shfmt runs on every staged shell file and WILL reformat your lib. It rewrote bt_audio.sh's { ...; } \ one-liners into multi-line blocks, which raised its denominator 99 -> 132 and invalidated a 100% measurement taken minutes earlier. Re-measure after pre-commit reformats anything, and put the post-format number in the evidence.
• bt_audio.sh is now exactly 250 lines, AT the cap. The next edit to it needs a file split, not a comment trim.
• A stub must consume what the real tool consumes. A bluetoothctl stub that read one line and exited closed the pipe, so the writing { ...; } | bluetoothctl block took SIGPIPE partway through and its later lines never ran. cat >/dev/null before exiting was worth 6 points.
• A function that appends to a global array must not be called inside $( ). The subshell dies with the substitution and the array is empty. Redirect to a file and cat it instead.
• A top-level stub function SHADOWS the subject's own for every later assertion. Put redefinitions in a subshell, and bind them with eval '_real_name() { _my_recorder "$@"; }' so shellcheck still sees the recorder as called (SC2329).
• A prepended stub dir cannot hide a real binary — use _t_hide.
• Do not pgrep -f shell_coverage_jail in a wait loop — the loop's own command line matches, so it never exits.
• --fail-on-case-error or you are flying blind. The jail suppresses the suite's stdout, so assertions fail invisibly without it.
• Commits exceed a 2-minute foreground timeout. Background them, and check the result — a file-length-cap failure aborted one commit that a task notification had reported as "completed (exit code 0)".
• mapfile is bash-only; the Bash tool's shell is zsh. Wrap in bash -c.
• Two-strike rule. Two failed attempts at the same thing -> stop, document, keep the working state.

Rules that will bite you

• No suppressions, ever. Fix SC2016/SC2034/SC2329/C901 at the source. The three _t_stub_* helpers and _quote_after() exist precisely because the alternative was a # shellcheck disable.
• Never weaken an assertion to move a number.
• Every commit touching code needs evidence in docs/superpowers/evidence/<slug>-<date>.json (validate with python3 meta/scripts/validate_evidence.py <file>), plus a contract in docs/superpowers/contracts/ once >=4 code files are staged.
• Put the MEASURED number in it, never a rounded or hoped-for one.
• New test files need the exec bit — chmod +x the working tree.
• Stage narrowly.
• Work directly on main. git stash and branch creation are blocked.

State as of 2026-08-22 (end of session)

Allowlist 93, down from 98 — five libs cleared. Nine commits, 4cfc3fdd..fa7b05da.

(Several commit messages say "95 -> 93" etc. and are one low: they counted total file lines instead of entries. The diff against 503ba1fc is the authority — five removals, 98 -> 93.)

lib	coverage	status
bt_report.sh	37/37 = 100.00%	CLEARED
bt_pairing.sh	65/65 = 100.00%	CLEARED
bt_audio.sh	127/127 = 100.00%	CLEARED
arch_perf_report.sh	59/59 = 100.00%	CLEARED
arch_perf_probes.sh	93/93 = 100.00%	CLEARED
bt_adapter.sh	68/69 = 98.55%	exempt, see artifact above

Also still exempt and already measured by an earlier session — do NOT redo: dwm_config.sh 98.48%, pacman_hook_stall_setup.sh 96.30%, aw_autostart.sh 97.56%, rpi_nc_install.sh 96.59%, transcribe_pkgmgr.sh 98.00%, clean_audio_filters.sh 91.40%, transcribe_deps.sh 92.31%.

rpi_nc_ca.sh (73/73) and dot_resolver_install.sh (39/39) are at 100% and could come off the allowlist whenever you want, by the three-step bar.

nc_php.sh does not measure at all — "kcov instrumented no lines", on the pre-fix instrument too. Not a regression; treat as unverified.

Not caused by this campaign

Python tests CI has failed since 2026-08-17 — pip hits resolution-too-deep on meta/requirements.txt. Excluded from check_ci_green.sh; do not fold it into this work.